← All papers
First page of Obstructions to Hadamard Deformations of Fourier Matrices

Obstructions to Hadamard Deformations of Fourier Matrices

Shuler Hopkins, Remus Nicoara

math.OA Nov 16, 2026 · v1 math.CO
The authors state the manuscript's results were formalized in Lean 4 with assistance from ChatGPT Astra; no Lean repository is cited.
We study deformations of the Fourier spin model commuting square in the space of commuting squares, or equivalently deformations of the matrix $\sqrt nF_n$, where $F_n$ denotes the Fourier matrix, by complex Hadamard matrices. If $n$ is neither a prime power nor twice a prime, we show that there exist unit vectors in the enveloping tangent space that cannot be realized as directions of convergence of sequential deformations. Consequently, the derivative of any differentiable Hadamard family through $\sqrt nF_n$ has rank strictly smaller than the dimension of the enveloping tangent space at $\sqrt nF_n$ (i.e., the defect of $F_n$). This is particularly surprising because for every $n$ there exist enough one-parameter analytic families to provide a basis of directions in the enveloping tangent space. We obtain cubic obstructions when $n$ has at least three distinct prime divisors, quartic obstructions when it has exactly two and a repeated factor, and quintic obstructions when it is the product of two distinct odd primes. These results extend our order $30$ result from \cite{HN}.

The paper asks whether every direction in the enveloping tangent space of complex Hadamard matrices at the scaled Fourier matrix √n F_n can be realized by sequential deformations. Equivalently, it asks whether such directions arise from deformations of the Fourier spin model commuting square.

The Fourier commuting square is decomposed via the Chinese remainder theorem into a tensor product, with a Weyl-type basis E_{x,y}. Using the holomorphic implicit function theorem, the non-tangent component of a deformation is expressed in terms of its tangent component. This yields third-, fourth- and fifth-order necessary conditions on directions of convergence. Explicit tangent vectors are then constructed for which these conditions fail.

If n is neither a prime power nor twice a prime, some unit tangent directions are not directions of convergence. The obstructions are cubic for three or more distinct primes, quartic for two primes with a repeated factor, and quintic for products of two distinct odd primes. Consequently, any differentiable Hadamard family through √n F_n has derivative rank at most d_n − 1; for example, the dephased rank at n=15 is at most 15 versus a defect of 16.